static void _image_orient(struct image *source,
struct object *o[5],
struct image *img[5])
{
int i;
struct { int x,y; } or[4]={ {1,0}, {1,1}, {0,1}, {-1,1} };
int x,y;
for (i=0; i<5; i++)
{
push_int(source->xsize);
push_int(source->ysize);
o[i]=clone_object(image_program,2);
img[i]=get_storage(o[i],image_program);
push_object(o[i]);
}
THREADS_ALLOW();
CHRONO("start");
for (i=0; i<4; i++) /* four directions */
{
rgb_group *d=img[i]->img;
rgb_group *s=source->img;
int xz=source->xsize;
int yz=source->ysize;
int xd=or[i].x;
int yd=or[i].y;
for(x=1; x<xz-1; x++)
for(y=1; y<yz-1; y++)
{
#define FOOBAR(CO) \
d[x+y*xz].CO \
= \
(COLORTYPE) \
my_abs( s[(x+xd)+(y+yd)*xz].CO - s[(x-xd)+(y-yd)*xz].CO )
FOOBAR(r);
FOOBAR(g);
FOOBAR(b);
#undef FOOBAR
}
}
CHRONO("end");
THREADS_DISALLOW();
}
// --------------------------------------------
void bench_mandelbrot_SIMD(int n, int max_iter)
// --------------------------------------------
{
// ne rien modifier dans cette fonction
int h, w;
float a0, a1, b0, b1;
vuint32 **M32;
uint32 **wM32;
uint8 **M8;
// chronometrie
int iter, niter = 4;
int run, nrun = 5;
double t0, t1, dt, tmin, t;
double cycles;
//puts("---------------------------");
//puts("-- bench_mandelbrot_SIMD --");
//puts("---------------------------");
h = w = n;
M32 = vui32matrix(0, h-1, 0, w/4-1);
M8 = ui8matrix(0, h-1, 0, w-1);
wM32 = (uint32**) M32;
// ne pas changer
a0 = -1.5;
a1 = +0.5;
b0 = -1.0;
b1 = +1.0;
CHRONO(calc_mandelbrot_SIMD_F32(M32, h, w, a0, a1, b0, b1, max_iter), cycles);
printf("SIMD F32: %10.2f\n", cycles/(n*n));
CHRONO(calc_mandelbrot_SIMD_I32(M32, h, w, a0, a1, b0, b1, max_iter), cycles); // facultatif
printf("SIMD I32: %10.2f\n\n", cycles/(n*n));
DEBUG(convert_ui32matrix_ui8matrix(wM32, 0, h-1, 0, w-1, M8));
DEBUG(SavePGM_ui8matrix(M8, 0, h-1, 0, w-1, "M_v.pgm"));
free_vui32matrix(M32, 0, h-1, 0, w/4-1);
free_ui8matrix(M8, 0, h-1, 0, w-1);
}
// ----------------------------------------------
void bench_mandelbrot_scalar(int n, int max_iter)
// ----------------------------------------------
{
// ne rien modifier dans cette fonction
int h, w;
int i0, i1, j0, j1;
float a0, a1, b0, b1;
uint32 **M32;
uint8 **M8;
// chronometrie
int iter, niter = 4;
int run, nrun = 5;
double t0, t1, dt, tmin, t;
double cycles;
//puts("-----------------------------");
//puts("-- bench_mandelbrot_scalar --");
//puts("-----------------------------");
h = w = n;
M32 = ui32matrix(0, h-1, 0, w-1);
M8 = ui8matrix(0, h-1, 0, w-1);
a0 = -1.5;
a1 = +0.5;
b0 = -1.0;
b1 = +1.0;
CHRONO(calc_mandelbrot_scalar(M32, h, w, a0, a1, b0, b1, max_iter), cycles);
printf("scalar: %10.2f\n", cycles/(n*n));
DEBUG(convert_ui32matrix_ui8matrix(M32, 0, h-1, 0, w-1, M8));
DEBUG(SavePGM_ui8matrix(M8, 0, h-1, 0, w-1, "M_scalar.pgm"));
free_ui32matrix(M32, 0, h-1, 0, w-1);
free_ui8matrix(M8, 0, h-1, 0, w-1);
}
void image_orient(INT32 args)
{
struct object *o[5];
struct image *img[5],*this,*img1;
int n;
rgb_group *d,*s1,*s2,*s3,*s0;
double mag;
int i, w, h;
CHECK_INIT();
this=THIS;
if (args)
{
if (TYPEOF(sp[-args]) == T_INT)
mag=sp[-args].u.integer;
else if (TYPEOF(sp[-args]) == T_FLOAT)
mag=sp[-args].u.float_number;
else {
SIMPLE_ARG_TYPE_ERROR("orient",1,"int|float");
UNREACHABLE(mag=0.0);
}
}
else mag=1.0;
if (args==1)
pop_n_elems(args);
if (args>1)
{
if (TYPEOF(sp[1-args]) != T_ARRAY)
SIMPLE_ARG_TYPE_ERROR("orient",2,"array");
if (sp[1-args].u.array->size!=4)
Pike_error("The array given as argument 2 to orient do not have size 4\n");
for(i=0; i<4; i++)
if ((TYPEOF(sp[1-args].u.array->item[i]) != T_OBJECT) ||
(!(sp[1-args].u.array->item[i].u.object)) ||
(sp[1-args].u.array->item[i].u.object->prog!=image_program))
Pike_error("The array given as argument 2 to orient do not contain images\n");
img1=(struct image*)sp[1-args].u.array->item[0].u.object->storage;
w=this->xsize;
h=this->ysize;
for(i=0; i<4; i++)
{
img1=(struct image*)sp[1-args].u.array->item[i].u.object->storage;
if ((img1->xsize!=w)||
(img1->ysize!=h))
Pike_error("The images in the array given as argument 2 to orient have different sizes\n");
}
for(i=0; i<4; i++)
img[i]=get_storage(sp[1-args].u.array->item[i].u.object,image_program);
pop_n_elems(args);
push_int(this->xsize);
push_int(this->ysize);
o[4]=clone_object(image_program,2);
img[4]=get_storage(o[4],image_program);
push_object(o[4]);
w=1;
}
else
{
_image_orient(this,o,img);
w=0;
}
s0=img[0]->img;
s1=img[1]->img;
s2=img[2]->img;
s3=img[3]->img;
d=img[4]->img;
THREADS_ALLOW();
CHRONO("begin hsv...");
n=this->xsize*this->ysize;
while (n--)
{
/* Första färg, sista mörkhet */
double j=(s0->r+s0->g+s0->b-s2->r-s2->g-s2->b)/3.0;
/* riktning - - riktning | */
double h=(s1->r+s1->g+s1->b-s3->r-s3->g-s3->b)/3.0;
/* riktning \ - riktning / */
int z,w;
if (my_abs((int)h) > my_abs((int)j))
if (h) {
z = -(int)(32*(j/h)+(h>0)*128+64);
w = my_abs((int)h);
}
else z=0,w=0;
else {
if (j) {
z = -(int)(-32*(h/j)+(j>0)*128+128);
w = my_abs((int)j);
}
else z=0,w=0;
}
d->r=(COLORTYPE)z;
d->g=255;
d->b = MINIMUM((COLORTYPE)(w*mag), 255);
d++;
s0++;
s1++;
s2++;
s3++;
}
CHRONO("end hsv...");
THREADS_DISALLOW();
if (!w)
{
add_ref(o[4]);
pop_n_elems(5);
push_object(o[4]);
}
}
